LECTURE 4-1 Common Sensing Techniques for Reactive Robots Introduction to AI Robotics (Sec. 6.1 6.5) 1
Quote of the Week Just as some newborn race of superintelligent robots are about to consume all humanity, our dear old species will likely be saved by a Windows crash. The poor robots will linger pathetically, begging us to reboot them, even though they'll know it would do no good. -Anonymous 2
ANNOUNCEMENTS Lab 4 - Line Following (PI Control) is due on Thursday, 4/1/10 The lab memo is due on Angel by midnight on Thursday, 4/1/10 Quiz 7 on Sec. 6.1 6.5, Lecture 4-1 on Tuesday, 3/30/10 3
OBJECTIVES Upon completion of this lecture the student should be able to: Define the difference between active and passive sensors Define the following terms in one or two sentences: proprioception, extereoception, exproprioception, proximity sensor, logical sensor, false positive, false negative List the metrics for rating sensors Describe the problems of specular reflection, cross talk, and foreshortening with an ultrasonic transducers Describe the types of behavioral sensor fusion and be able to apply to a real world problem Write perceptual schemas from any logical equivalent range sensor to produce a polar plot for obstacle avoidance behavior 4
PERCEPTION One of the most important tasks of an autonomous mobile robot is perception. Perception is used for the robot to acquire knowledge from it s environment. Perception involves taking measurements using various sensors and extracting meaningful information. Localization "Position" Global Map Cognition Environment Model Local Map Path Perception Real World Environment Motion Control 5
PERCEPTION Perception in a reactive robot system has two roles: Release a behavior Guide the actions of the behavior All sensing is behavior-specific The sensor or transducer is a device that measures the attributes of the world A transducer is the mechanism of a sensor that transform the energy associated with what is measured into another form of energy. (i.e. sound, light, pressure, temperature to an analog or digital form) 6
SENSING A robot s intelligence depends on The quality and quantity of its sensors The ability to process and speed of processing sensory input Sensing allows a robot to know its state or description of itself at any point in time A robot s state may be visible, partially hidden, or hidden A state may be discrete or continuous A state space consists of all of the possible states a system can be in 7
SENSING CONT. External state is the robot s perception of the world Internal state is the robot s perception of itself Representation or internal model is created when a robot uses its internal state to remember information about the world. A robot s sensor space or perceptual space is the space of all possible sensory readings based upon all of the robot s sensors 8
A MODEL OF SENSING 9
SENSING IN REACTIVE PARADIGM 10
ACTIVE VS PASSIVE SENSING Active Sensors Emits some form of energy and then measures the return to understand the environment (sonar, laser) Passive Sensors Receives energy already in the environment (camera) Thermal sensor Stereo Camera pair Laser ranger Sonars Bump sensor 11
SIMPLE VS. COMPLEX SENSORS Simple sensors do not require a great deal of processing or computation but the information they provide is simple or limited (i.e. light levels, presence or absence of objects, distance to objects) Passive sensors can be simple or complex the camera is a complex passive sensor Active sensors are not necessarily complex break beam sensors are simple active sensors Whether a sensor is simple or complex is determined by the amount of processing its data require Whether a sensor is active or passive is determined by the mechanism used to extract data 12
SENSORS A robot s intelligence is typically measured by the diversity of its sensor suite. Proprioceptive Sensors perceive elements in the robot s internal state. For example, encoders on motors for distance or velocity or battery voltage. Exteroceptive Sensors perceive elements in the state of the external world around the robot. For example, infrared sensors for distance, temperature or light intensity. Proprioceptive and Exteroceptive sensors combine to make the robot s perceptual system 13 ECE497 Lecture 4-1: Common Sensing Techniques for Reactive Robots (C.A. Berry)
SENSOR DIVERSITY Robots have different types of sensors to allow: Flexibility in type of data (direction, distance, light, sound, temperature) Sensor fusion to obtain a more accurate representation of the world Multiple Sensors Speed up the rate of environment readings Provide redundancy and fault tolerance Save power 14
SIGNAL VERSUS STATE Sensors do not provide state information but rather data or physical quantities. These measurements must be processed in order to be useful Because sensors are prone to error, there are several sensors that measure the same quantity and can be used for sensor redundancy Sensor fusion is when multiple sensors are used to create some knowledge or get better information about the robot s internal and/or external state. 15 ECE497 Lecture 4-1: Common Sensing Techniques for Reactive Robots (C.A. Berry)
SENSOR FUSION To account for inaccuracies, multiple sensors are often combined (or fused) Sensor Fusion combines sensor readings from The same sensor Taken as an average, minimum or maximum over some small time interval (infrared sensor) Multiple similar sensors Individual sensors read from different directions (sonar ring) Different kinds of sensors Combine sonar, infrared, and vision measurements 16
BEHAVIORAL SENSOR FUSION Three basic combinations of sensors Redundant (or competing) Complementary Coordinated When a sensor leads to robot to believe a percept is present but it is not is called a false positive When a robot misses a percept it is a false negative Redundant sensors return the same percept (physical or logical) Complementary sensors return disjoint types of information about a percept Coordinated sensors use a sequence of sensors for providing focus of attention 17
UNCERTAINTY Uncertainty refers to the robot s inability to be certain about the state of itself or the environment. Some of the sources of uncertainty are: Sensor noise and errors Sensor limitations Effector and actuator noise and errors Hidden and partially observable state Lack or prior knowledge about the environment Dynamic and changing environment 18
BEHAVIORAL SENSOR FUSION: Sensor fission Emergent behavior (competing) Action-oriented sensor fusion (complementary) Sensor fashion (coordinated) 19
SENSOR QUESTIONS Given a sensory reading, what should I do? Easy to answer with a simple sensor, i.e. if the bump sensor is triggered, the robot should stop Hard to answer with a complex sensor, i.e. if the camera image contains a wall, the robot should? Given a sensory reading, what was the world like when the reading was taken? May not be easy to answer with simple or complex sensors Simple sensors do not prove enough information for reconstruction of the world. i.e. if the bump sensor is triggered, what did it hit? Complex sensors provide this information but require a great deal more processing 20
HOW DO YOU RATE SENSORS? Field of view, range: does it cover the right area Accuracy & repeatability: how well does it work? Responsiveness in target domain: how well does it work for this domain? Power consumption: may suck the batteries dry too fast Reliability: can be a bit flakey, vulnerable Size: always a concern! Computational Complexity: can you process it fast enough? Interpretation Reliability: do you believe what it s saying? 21
PROXIMITY SENSORS Detect objects within a specific range from the robot Provide a binary signal according to some threshold Tactile sensors are an example of proximity sensors Non-tactile sensors that detect the absence or presence of a light reflect are encoders Non-tactile sensors are usually active Range sensors can be configured as proximity sensors by setting a threshold 22
ACTIVE RANGING Active ranging sensors are the most popular sensors in mobile robotics Active ranging sensors are used for Obstacle detection Obstacle avoidance Localization Environment modeling Ultrasonic sensors and laser range sensors use the propagation speed of sound or electromagnetic waves. 23
COMMON RANGE SENSORS Commonly used range sensors in robotics include: Tactile and proximity sensors Ultrasonic sensors Infrared range sensors Laser range finders Vision systems Each varies in complexity, size, weight, expense, accuracy, etc.. The detection range is defined as the maximum distance that the sensor can read reliably 24
INFRARED RANGE SENSORS Emit light from Infrared LED Light is reflected from object Receiver measures strength of light returned Range depends on object properties Shiny objects (metal) are difficult to detect Cannot detect glass White/black surfaces report different ranges 25
TIME OF FLIGHT time of flight is used to determine distance to objects The measured pulses typically come form ultrasonic, RF and optical energy sources. d = c * t d = round-trip distance c = speed of wave propagation t = time of flight 26
INFRARED SENSORS: MAPPING EXAMPLE Use the IR sensor to compute the range to obstacles along a long a path in the following environment 27
INFRARED SENSORS: MAPPING EXAMPLE Take measurements along the path at particular locations 28
INFRARED SENSORS: MAPPING EXAMPLE Blue lines show readings to obstacles form the robot s center position (x, y) 29
INFRARED SENSORS: MAPPING EXAMPLE The resulting map has reasonable accuracy The map can be refined by taking additional readings 30
IR SENSORS: ADVANTAGES AND DISADVANTAGES Advantages Reliable with good precision Small beam angle inexpensive Disadvantages Sensitive to smoothness Sensitive to angle to obstacles Short range Prone to interference from ambient light Cannot detect glass, mirror, shiny surfaces 31
SONAR Ultrasound (sonar) refers to the range of frequencies of sound that are beyond human hearing The process of finding your location based upon sonar is echolocation Sonar are active sensors that emit a chirp or ping and use time of flight to determine distance The transducer on a sonar emits the chirp/ping and receives the sound (echo) that comes back. Mechanical energy is converted into sound as the membrane on the transducer flexes to produce a ping ECE497: Introduction to Mobile Robotics (C.A. Berry) - Complex Sensors
SONAR: RELIABILITY Blind zone is when an echo arrives before the transducer is ready to receive and objects are not detected reliably Sensor readings vary based upon: Distance to object(s) Angle that object makes with respect to sensor axis Direction that objects enter sensing range ECE497: Introduction to Mobile Robotics (C.A. Berry) - Complex Sensors
SONAR: SPECULAR REFLECTION Specular reflection can cause reflected sound to Never return to the transducer Return to the transducer too late The results is that the distance measurement is too large and inaccurate ECE497: Introduction to Mobile Robotics (C.A. Berry) - Complex Sensors
SONAR: SENSITIVITY Sensitivity to obstacle angle can result in improper range readings When the beam angle of incidence falls below a certain critical angle specular reflection errors occur ECE497: Introduction to Mobile Robotics (C.A. Berry) - Complex Sensors
SONAR: RESOLUTION Distance and angular resolution decreases as objects become further from the sensor Multiple close objects cannot be distinguished Gaps such as doorways cannot be detected ECE497: Introduction to Mobile Robotics (C.A. Berry) - Complex Sensors
ECE497: Introduction to Mobile Robotics (C.A. Berry) - Complex Sensors ULTRASONIC RANGE SENSORS: REDUNDANCY To increase beam width (resolution), two sensors are used together Detection in either or both sensors allows for increased resolution
ULTRASONIC RANGE SENSORS: APPLICATIONS Distance Measurement Mapping: Rotating proximity scans (maps the proximity of objects surrounding the robot) Scanning at an angle of 15º apart can achieve best results Doorway chair Robot Length of Echo chair Scan moving from left to right ECE497: Introduction to Mobile Robotics (C.A. Berry) - Complex Sensors
ULTRASONIC RANGE SENSORS: MAPPING To perform mapping take multiple readings: Rotate the sensors Rotate the robot chassis Use multiple sensors at fixed positions on chassis ECE497: Introduction to Mobile Robotics (C.A. Berry) - Complex Sensors
ECE497: Introduction to Mobile Robotics (C.A. Berry) - Complex Sensors ULTRASONIC RANGE SENSORS: MAPPING EXAMPLE Use sonar mounted to the front of a robot to compute the ranges to obstacles from a location in the environment
ULTRASONIC RANGE SENSORS: MAPPING EXAMPLE Blue lines show sonar readings detected from the robot s position (x, y) to the obstacle position (x o, y o ) ECE497: Introduction to Mobile Robotics (C.A. Berry) - Complex Sensors
ULTRASONIC RANGE SENSORS: MAPPING EXAMPLE The sonar data produces a rough outline of the environment with some inaccurate readings ECE497: Introduction to Mobile Robotics (C.A. Berry) - Complex Sensors
ULTRASONIC RANGE SENSORS: CROSSTALK Using multiple fixed sensors can lead to crosstalk Crosstalk is interference in which echoes emitted from one sensor are detected by others ECE497: Introduction to Mobile Robotics (C.A. Berry) - Complex Sensors
ECE497: Introduction to Mobile Robotics (C.A. Berry) - Complex Sensors ULTRASONIC RANGE SENSORS: CROSSTALK SOLUTION Crosstalk signals are impossible to detect unless signals are unique (coded) Crosstalk can be reduced by carefully timing the emitting of signals Emit from one and wait for a time interval Emit from a select few that may not have interference Emit adjacent sensors at different frequencies
ULTRASONIC RANGE SENSORS: NOISE ISSUES ECE497: Introduction to Mobile Robotics (C.A. Berry) - Complex Sensors
ULTRASONIC RANGE SENSORS: ADVANTAGES AND DISADVANTAGES Advantages Reliable with good precision Not as prone to outside interference Good maximum range Inexpensive Disadvantages Sensitive to smoothness Sensitive to angle to obstacles (specular reflection) Poor resolution Prone to self-interference from echoes Cannot detect obstacles too close Soft surfaces absorb sound energy Bandwidth ECE497: Introduction to Mobile Robotics (C.A. Berry) - Complex Sensors